Understanding, modeling, prediction or forecasting of precipitation has always been important. Precipitation is the driven factor of hydrological cycle and is one of the main sources of water without which the humankind cannot survive. In particular, it becomes the unique source of water in arid regions where surface water courses are generally intermittent or ephemeral in nature. In such regions the main water income is from the groundwater storages that are again fed by precipitation. In such regions, therefore, precipitation becomes much more interesting than runoff to analyze as flow records are generally insufficient in length to obtain accurate estimations. Analysis of precipitation is evident not only because of that reason but also for agricultural and socio-economical activities, for increasing human and environmental demands as well as for planning and management of water resources. Precipitation changes both in time and space and affects on other components of the hydrological cycle; i.e. surface runoff, infiltration, groundwater, seepage, percolation, evaporation, transpiration, etc. Time variation comes from the seasonal clima-tological changes in the atmosphere whereas spatial change is due to the topographical heterogeneity on the earth surface. Precipitation amount should be predicted accurately for an affective water resources management and planning. Prediction of hydrometeorological time series is difficult because of uncertainty in the parameters which affect the time series. In literature, the applications of artificial neural networks to forecast arid-region precipitation are limited. Forecasting monthly precipitation in arid regions is investigated by means of the conditional feed-forward backpropagation artificial neural network combined with Markov chain. Amman meteorological station from the hilly region in Jordan is selected for application. Various homogeneity tests are used for the data, which is found homogeneous according to the results of the tests. The feed-forward backpropagation artificial neural networks, statistical, and stochastic methods provided negative approximations for some of the low precipitation (dry months) whereas Markov chain forecasts are bounded with zero as the lower limit. Prediction using Markov chain does not generate physically unexplainable results mainly because the forecasts of the feed-forward backpropagation combined with Markov chain method are bounded with the minimum (zero) precipitation. Antecedent monthly precipitation data are used as input to the model to predict the total precipitation in the next month. Selection of the inputs is based on the correlation coefficient. For the station analyzed, it is observed that precipitation in any month was highly correlated to its previous two months and its previous year record. A trial-and-error based construction for various artificial neural networks are tested. A total of 28 combinations of input variables are investigated. It is finally ended with an architecture that uses, as input variables, precipitation in two previous months and the previous year's precipitation of the month to be forecasted. Also a periodical component is added into the input vector to simulate the existing periodicity in the monthly precipitation. Since the sigmoid function is used as the activation function for hidden and output layers, the model input and output are scaled appropriately to fall within the function limit (zero to one). Conditional feed-forward backpropagation artificial neural network combined with Markov chain is trained using the Levenberg-Marquardt training algorithm. After training is over, the weights are used to test the network performance on the test data. In conclusion, it is seen that the conditional model considerably improves the accuracy of the one month ahead precipitation forecasting compared to the unconditional model.